Social Network Forensics: Tapping the Data Pool of Social Networks
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Social Network Forensics:
Tapping the Data Pool of Social Networks
Martin Mulazzani, Markus Huber and Edgar Weippl
Abstract
With hundreds of millions social network users worldwide, forensic data
extraction on social networks has become an important research problem.
The forensic data collection is however tightly connected to the service op-
erator, which leads to data completeness and presentation format issues.
Online social networks imply that all user communication is stored entirely
at the service operator, without direct access for investigators. In this paper
we identify important data sources and analytical methods for automated
forensic analysis on social network user data. We furthermore show how
these data sources can be evaluated in an automated fashion, and without
the need of collaboration from the social network operator. While our pro-
posed methods apply to the great majority of social networks, we show the
feasibility of our approach on basis of Facebook.
Keywords: Online forensics, social networks, visualization1 Introduction
With the increasing usage of social networks and the emerge of cloud com-
puting, digital forensics faces novel research problems and challenges. The
number of users of these services increases steadily, with e.g. Facebook cur-
rently claiming to have 800 million users [4]. While traditional forensics relies
on the physical acquisition of hardware [13, 14] and the usage of hashsums to
ensure evidence reliability, this approach does not scale to cloud services and
their use of distributed datacenters. With the lack of standardized forensics
APIs as well as unified processes for service operators, isolated solutions are
still in widespread use. Another important aspect of forensics is the proper
visualization of data [22, 16] due to the vast amount of available data.
It is furthermore hard to visualize gathered social networking data in
a way that can answer common questions of interest on a first sight, so
that people without technical background can understand it. This has been
shown for example in the case of the consolidated.db from Apple’s iPhone:
the file contained geolocation information which has been already outlined
in 2010 [21]. However, the consolidated.db problem got widespread attention
with the release of the iPhoneTracker software [6] in April 2011, which visu-
alized the collected data. Due to the iPhoneTracker software, Apple finally
had to review and change their data collection process[1].In this paper we identify data sources of interest for forensic examinations
on social networks, and how they can be leveraged in an automated fashion.
We furthermore identify graphs of interest that can be generated from these
data sources and can answer many possible questions of a forensic examiner
on first sight. To the best of our knowledge there has not yet been any work
on forensic analysis of data from social networks without the collaboration of
the social network operator. We show example graphs and present possible
visualizations, which can and should be used for social network analysis and
will be released under an open source license.
The rest of the paper is organized as follows: Section 2 gives an overview
of social networks, and how they are already used for conducting and solving
crimes. Section 3 shows what data can be extracted from social networks for
social network forensic investigation, while Section 4 explains the data and
how it can be visualized. We show the feasibility of our approach in Section 5
and conclude in Section 6.
2 Background
Social network forensics has to rely on a limited set of data sources in many
cases. Acquiring the server’s hard drives is not feasible, and leveraging the
service operator’s data directly requires the service operator’s cooperation.
If at all, the investigator can submit requests to the operator and may ormay not receive all the relevant data (e.g. written in the Facebook law en- forcement guidelines published by the EFF [3]). This is in clear contradiction to the guidelines for evidence collection [13], as the investigator is unable to show that the evidence is authentic, complete, and reliable. Network forensic frameworks like PyFlag [15] and Xplico [8] simply cannot see or access all the data, as they are solely passive. Our approach, on the other hand, does not require the cooperation of the social network operator, and can ensure these properties due to the open nature of our collection methodology and tools. 2.1 Data Acquisition Before being able to analyze social network data the data has to be gathered and acquired. While traditional forensic methods can be used to extract artifacts from local webbrowser cache [2], numerous other ways are possi- ble on the communication layer. These range from passive sniffing on the network to active attacks like sniffing on unencrypted Wifis [20] or in com- bination with ARP spoofing on LANs. The recently proposed friend in the middle attack [19], which uses a third party extension for the social network in combination with a traditional crawler component, could be used as well. Crawling however is limited, as metadata and accurate timestamps are not shown on webpages. They are only available by using the social network APIs, which extend the available data of the webinterface. Even though it would be possible to use passive logging on the communication layer, for
example in cases where a judge ordered lawful interception on the Internet
connection of a suspect, this approach is limited as well as it would take a
tremendous amount of time for collecting information, and completeness is
hardly possible. Furthermore, many social networks offer the possibility to
encrypt data on the communication layer by using HTTPS, rendering passive
attacks useless.
While Facebook announced recently that users are now able to download
all their profile data, the data provided by our method is far superior com-
pared to the Facebook profile download option which lacks e.g., important
metadata and is thus not useful for forensics. In general, it is not possible for
a user to download everything that is connected to his or her profile on the
social network. Another interesting feature recently announced is Facebook
Timeline, which encourages users to never delete anything from the social
network, and to use it as an historic archive. This surely is interesting for
forensic examinations, as the user is less likely to delete data.
3 Social Network Data Pool
While social networks vary in features and architecture, we identify the fol-
lowing generic data sources to be of interest in forensic examinations on social
networks:
• The social footprint: What is the social graph of the user, with whomis he or she connected (“friend”)?
• Communications pattern: How is the network used for communicating,
what method is used, and with whom is the user communicating?
• Pictures and videos: What pictures and videos were uploaded by the
user, on which other peoples pictures is he or she tagged?
• Times of activity: When is a specific user connected to the social net-
work, when exactly did a specific activity of interest took place?
• Apps: What apps is the user using, what is their purpose, and what
information can be inferred in the social context?
All this information cannot be found on a suspect’s hard drive, as it is
solely stored at the social network operator. Especially for people that use
the social network on a daily base a plethora of information is stored at the
social network operator. Facebook claims that more than 50% of its users
use it any given day, which would be something around 400 million users [4].
Sometimes information is cached locally, but this is not a reliable source of
information as it is neither complete nor stored persistently. Depending on
the implementation of the social network, the availability of data itself and
the possibility to retrieve the data via API calls can vary among different
social networks. However, most of this data can be extracted either directly,
or inferred with low overhead without the collaboration of the social network
operator. Once the data is available to the investigator, the full spectrum ofsocial network analysis can be conducted [23]. The easiest way for obtaining the data is of course with the consent of the user, who can provide username and password. While the data can be easily analyzed manually afterwards to answer specific questions, the humongous amount of data requires automated tools for an forensic examiner to see the full picture. 4 Visualizations Based on the social network data pool we define the following graphs of interest and visualizations for social network forensics. 4.1 Basic Visualizations Social Interconnection Graph: It is trivial to retrieve the list of friends from social networks. In most social networks this is public information, or can be easily collected [12] even without entering the social circle of the ac- count under investigation. However, it is not trivial to cluster these friends, namely to find out who is connected with whom: is a specific contact part of the cluster of work-friends, or are they directly related? In our approach we use a feature of the Facebook API which allows an ap- plication to query if two users are connected. This allows our software to cluster the friends of a user into different groups e.g., people from work, school, family and more, as the members of the groups are much more likely to know each other. The graph can be represented as an undirected graph
G =< V, E > where V = v1 , v2 , ...vn is the set of friends of a user, and E = (vx , vy ), ... is the set of edges that connects two nodes in case they are friends on the social network. Highly connected nodes have a high degree, representing well connected friends that know most of the suspect’s friends as well. An example graph will be shown in Section 5. Social Interaction Graph: For many investigations it is of importance to find out who communicated with whom. Various ways of communication are possible among users, like wall posts, direct messages, group communica- tion or following public announcements. Communication can be represented as a directed graph G =< V, E > where the nodes V = v1 , v2 , ...vn are all the friends while the edges E = (vx , vy ), ... are directed and the weight of (vx , vy ) is incremented for every message sent from vx to vy . In this paper, however, we do not distinct between the different forms: direct messages, wall posts, and tags are treated equally as these are the most direct form of communi- cation. We leave finding different metrics for future work which would allow the investigator to add custom weights to specific communication forms. An investigator can easily identify the top communication partners on a first sight, and compare them with e.g., obtained phone records. An example graph generated by our tools will be shown in Section 5 as well. Complete Timeline: With social network users being online 24/7 by using mobile clients on smartphones, the timeline becomes of increasing im-
portance. Not only activity of the user itself can be extracted, but also the activity of all his or her friends. Often the times of activity can be seen easily, if properly visualized. To make analysis feasible it is necessary to use respectively allow different data layers: activity of the user, the friends, group activities, reactions on events from friends, and so forth. It is also crucial that the timeline is zoomable, to visualize time ranges of importance - a single day can have easily more than 500 events for a given profile and all his or her friends. Location Visualization: Geotagging and location applications are novel features with increasing usage that needs to be reflected in forensic exami- nations. With foursquare [5] and Facebook Places, just to name a few, the geolocation information stored in social networks is growing steadily. While our toolset is not yet available to extract geodata, we believe that this will become more and more of an issue. Digital cameras as well as smartphones automatically geotag pictures taken with the exact location. Up till now, most social networks remove metadata during the transformation for picture storage [10], but this might change in the future. 4.2 Advanced Visualizations & Information Inference While the features discussed so far are rather straightforward, we believe that the following list of advanced features and components should become
standard tools in forensics: Event tracking: For viral scammers and other malicious applications that use the social network for propagation it might be of interest who or what started such a series of events. Tracking such events is not straightfor- ward, but with a collection of social network footprints of various users these events can be easily dissected. This would allow insight into dissemination characteristics and propagation tactics of scammers, as well as advanced an- alytical capabilities. Timeline matching: In highly centralized systems such as online social networks, an investigator has the benefit of consistent timestamps as they are provided by the social network. The operators often run their won NTP infrastructure, and keep the clocks consistent across thousands of servers. This can then be used to match timelines of different profiles, and eventu- ally create an exact timeline for a complete cluster of friends or even bigger. While this has been proposed recently for the NTFS file system [17], we be- lieve that this will be of importance for social networks and cloud computing as well. Differential Snapshots: Once a forensic image of a user profile is col- lected, at a later point in time the image might look completely different. Therefore, the forensic framework must provide the functionality to not only
visualize the social network data of a user, but also the functionality to vi- sualize differences with previous images of the same user. 5 Results We implemented the data collection methodology outlined in [18] to collect data from Facebook, currently the biggest social network service. We then parsed the output and generated the graphs as seen below. Data acquisi- tion takes approximately 20 minutes per account, which is in our opinion a reasonable amount of time for data collection. The data currently includes all the social connections, direct communication, pictures and much, much more. 5.1 Results Visualization For the social interconnection graph we used a feature from the Facebook API that allows an application to query if two specific users are connected. We iteratively tested if the first friend is in connection with the n − 1 friends of the tested profile, then tested for the second the remaining n − 2, and so forth. We then extracted the social interconnection graph and plotted it with Gephi [9], an open source graph visualization tool, using the Fruchterman- Reingold algorithm [11]. The nodes that seem to be from the same cluster are colored accordingly without manual intervention, which makes cluster analysis for a forensic examiner very convenient. An example of a social in-
terconnection graph from one of the authors can be seen in Figure 1. Please note that the names of the profiles have been replaced by a random subset of the list of computer scientists on Wikipedia [7]. As the replaced set is random, it obviously does not show real connetions between the computer scientists. With the data from Facebook it becomes possible to create different so- cial interaction graphs. In our implementation we created different graphs for different forms of interaction, while they could be easily integrated to a complete social interaction graph. An example for a social interaction graph based on tags in pictures on Facebook can be seen in Figure 2. The graph is created using the following steps: (1) starting from an account under inves- tigation, all pictures from all friends are collected, and searched for tagged people. (2) People that are tagged in pictures, and not in the list of friends, are ignored. (3) If the tagged person is in the friend list as well, an edge is added between the two nodes pointing from the profile that uploaded the picture to the profile that is tagged, or the weight increases by one if the edge already exists. The edges are directed and weightened. An investigator can find with this graph persons that have a tight social connection. Another form of social interaction graph can be created using direct mes- sages: instead of using picture tags, an edge is added between two nodes if the profile under investigation exchanged messages with the other profile.
William H. Press
Guy L. Steele, ...
Srinidhi Varada... Guido van Rossum
Daniel Mopati K...
Yukihiro Matsum...
G.M. Nijssen
Amir Pnueli
David H. D. War...
Terry Winograd
Gordon Plotkin Sophie Wilson Vinod Dham
James Z. Wang Michael I. Schw...
Adi Shamir
Barbara J. Grosz
Alan Perlis Joseph Weizenbaum Carl Kesselman
Susan Dumais
Philip-Emeagwali
Cleve Moler
Grady Booch
James G. Nell
Donald Firesmith
Donald Knuth
Janet L. KolodnerLarry Wall
Stephen Muggleton James Gosling
Frances E. Allen Alfred Aho
Jack E. Bresenham
Butler W. Lampson
Sanjeev Arora
James C. Beatty...
David A. Huffman Joseph F Traub
George
Jeffrey D. Ullman Boole Per Brinch Hansen François Vernadat
Ken Kennedy Wim Ebbinkhuijsen Richard Veryard
Chris McKinstry
Adriaan van Wij... Joseph Halpern
Jie Wu Thomas Sterling Gordon Cormack
Börje Langefors
Michael Stonebr... Les Hatton Leonard Adleman Herman Hollerith
Michael Garey Joseph Kruskal
Marvin Minsky
Kathleen R. McK...
Edward H. Short...
Bruce Schneier Andrew Herbert
Ken Thompson
Edwin Catmull
John George Kem...
Godfried Toussa... Edsger Dijkstra Fred B. Schneider
Thomas E. Kurtz Jonathan James Douglas McIlroy
Carl Sassenrath
Marilyn A. Walker He Jifeng
Robert Floyd Seinosuke Toda
Edgar F. Codd
Fernando J. Cor... Neil J. Gunther
James H. Wilkin...
Eric Horvitz
David Liddle
Christopher Ric... Bernard Galler
Herbert W. Franke Zvi Galil
Marco Dorigo Admiral Grace H...
Ron Rivest
Winston W. Royce
Luis von Ahn
William Kahan Frieder Nake
Tim Berners-Lee Mihai Nadin
John McCarthy Edmund M. Clarke
David S. Johnson
Gerrit Blaauw John L. Hennessy Jiawei Han
Ole-Johan Dahl
Charles E. Leis...
Tom Gruber
Bill Gropp
Leonard Kleinrock Robin Milner
Erik Demaine
Steve Whittaker Dennis E. Wisno... Lambert Meertens
Paul Dourish
Ivar Jacobson
Hector Garcia-M...
Joyce K. Reynolds John Koza
Alan Burns
Douglas Lenat Stephen R. Bourne David A. Bader
Stephen Wolfram
Yuri Matiyasevich Tom Lane (compu...
Paul Graham
Michael Dertouzos
Ian Goldberg
Peter Wegner
Michael L. Scott
Brian Cantwell ... J.C.R. Licklider
Martin Hellman
William Wulf
Andries van Dam Sjaak Brinkkemper
Simon Colton Jonathan Bowen Zhou Chaochen
Alan Turing Andrew Ng
James Martin
Andrew Appel Patrick Cousot Vinton Cerf
Kurt Gödel
Bert Bos
Murray Turoff
Christopher Str... Dragomir R. Radev
Douglas T. Ross
Joseph Sifakis
Mark Overmars David Gelernter
Amos Nuwasiima
George Sadowsky Alexander Dewdney
Alan Dix
Emil Post
Bertrand Meyer Adam Riese
Jeff Rulifson
John C. Reynolds
Andrey Ershov
Brian Randell
Manindra Agrawal
John Backus
Jan Weglarz
Gordon Moore
Figure 1: Anonymized Social Interconnection GraphYuri Matiyasevich
Marvin Minsky
Thomas Sterling
G.M. Nijssen
Kurt Gödel
Andries van Dam
David H. D. Warren Stephen Cole Kleene
George Novacky
David Wagner Michael L. Scott Ramesh Jain
Zvi Galil
Whitfield Diffie
Shmuel Winograd
David Karger
Michael O. Rabin
Charles Babbage
Yukihiro Matsumoto
James G. Nell
Alonzo Church Ken Kennedy
Paul Dourish
Dennis E. Wisnosky
Douglas Engelbart
Erik Demaine
Anthony James Barr
James Z. Wang
Janyce Wiebe
David A. Bader
Figure 2: Anonymized Social Interaction Graph using Picture Tags
Intuitively, an edge pointing to a node represents a message sent to that pro-
file. Again the edges are weightened, for the number of messages sent. An
example for a social interaction graph using direct message communication
can be seen in Figure 3.
Our data collection method also allows automated creation of timelines.
While this is currently under development, an example for a timeline of events
on Facebook over a 24-hour period can be seen in Figure 4.
5.2 Threats to Validity
Whilst our method is novel and can be easily used in addition to already ex-
isting and deployed social network analysis methods (i.e., subpoena requestsFigure 3: Social interaction graph using direct messages
John Doe 7:12:50 AM
12:00:50 PM 10:08:50 PM
Wall Post ID 123456789 Private Message
ID 11111111 Like Wall Post ID 00000000
of User 123456
Privacy: ALL_FRIENDS
Type: Link
6:24:50 PM
Uploaded digital picture
ID 00000001
To Javier Rodrigez
UTC-2 ID 77777777
Privacy: EVERYBODY
ID 555555
01:00 02:00 03:00 04:00 05:00 06:00 07:00 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00
9/14/2011 12:00 AM 9/15/2011 12:00 AM
Private Message
ID 00000000 User Max Mustermann ID 4444444 User Javier Rodrigez ID 5555555
To John Doe Comment Wall Post ID 123456789 Comment Wall Post ID 123456789
ID 1111111111 2:08:50 PM 3:44:50 PM
4:32:50 AM
Figure 4: Anonymized Example Timeline for 24-hour periodto the social network operator) they introduce new challenges at the same time as they solve others. One of the most obvious drawbacks is that the data collection is hardly reproduceable: the timelines and graphs generated will look differently for multiple runs as the social network is very dynamic in nature, and the amount of data rather big. Within a single day a user could change his social interconnection graph to large extend, and could try to hide his or her communication in covert traffic. Some other data, like login IP addresses as provided by the Facebook NeoPrint [3], are only available to the operator of the network and not accessible with our method, neither with an automated webbrowser or an API. Furthermore it is not easily possible to guarantee completeness: once data is deleted by the user it can only be undeleted from the social network operator, and is thus not available for our analysis. We believe, however, that most of the data items are rather static in nature, and that our methods are applicable and auxiliary for forensic investigations. 5.3 Future Work While some of the features are already implemented in our toolkit, we plan on implementing the missing visualizations as well as the automated report generation soon. We would also like to extend the number of supported social networks. It would be very interesting to analyze how static social graphs are, which is so far beyond the scope of this paper. This information could be used to confirm the validity of our methods even further.
6 Conclusion
Social networks and the cloud computing paradigm will undoubtedly change
the way forensics examinations are done in the near future. In this paper we
identified valuable data sources in social networks, how they can be lever-
aged for analysis, and discussed to what extend this is possible without the
collaboration of the social network operator. Compared to traditional so-
cial analysis this can be done fully automated and allows cluster analysis as
well as different timeline visualizations. We implemented a proof-of-concept
application for creating social interconnection and social interaction graphs
(with whom is a user connected, with whom does the user interact often)
based on Facebook to show the feasibility of our approach, which we will
release as open source software.
References
[1] Apple Promises Fix for Location-Gathering ‘Bug’ on iPhone.
Online at http://www.wired.com/gadgetlab/2011/04/
iphone-location-bug/.
[2] Facebook Artifact Parser, retrieved September 15th, 2011. Online at
http://trustedsignal.com/code/fbartiparse.py.
[3] Facebook Law Enforcement Guidelines, retrieved September 15th, 2011.
Online at http://www.eff.org/files/filenode/social_network/Facebook2010_SN_LEG-DOJ.PDF.
[4] Facebook Statistics, retrieved September 7th, 2011. Online at http:
//www.facebook.com/press/info.php?statistics.
[5] foursquare. Online at https://foursquare.com/.
[6] iPhoneTracker, retrieved September 7th, 2011. Online at http://
petewarden.github.com/iPhoneTracker/.
[7] List of computer scientists - Wikipedia, the free encyclopedia, retrieved
September 15th, 2011. Online at http://en.wikipedia.org/wiki/
List_of_computer_scientists.
[8] Xplico - Internet Traffic Decoder. Network Forensic Analysis Tool. On-
line at http://www.xplico.org.
[9] M. Bastian, S. Heymann, and M. Jacomy. Gephi: An open source soft-
ware for exploring and manipulating networks. In International AAAI
Conference on Weblogs and Social Media, pages 361–362, 2009.
[10] D. Beaver, S. Kumar, H. Li, J. Sobel, and P. Vajgel. Finding a needle in
haystack: Facebook’s photo storage. In Proceedings of the 9th USENIX
conference on Operating systems design and implementation, pages 1–8.
USENIX Association, 2010.[11] V. Blondel, J. Guillaume, R. Lambiotte, and E. Lefebvre. Fast unfold-
ing of communities in large networks. Journal of Statistical Mechanics:
Theory and Experiment, 2008:P10008, 2008.
[12] J. Bonneau, J. Anderson, R. Anderson, and F. Stajano. Eight friends are
enough: social graph approximation via public listings. In Proceedings of
the Second ACM EuroSys Workshop on Social Network Systems, pages
13–18. ACM, 2009.
[13] D. Brezinski and T. Killalea. Rfc 3227: Guidelines for evidence collection
and archiving, 2002. Online at www.faqs.org/rfcs/rfc3227.html.
[14] B. Carrier. File System Forensic Analysis. Addison-Wesley Professional,
2005.
[15] M. Cohen. Pyflag-an advanced network forensic framework. digital
investigation, 5:S112–S120, 2008.
[16] G. Conti. Security Data Visualization: Graphical Techniques for Net-
work Analysis. No Starch Pr, 2007.
[17] X. Ding and H. Zou. Time based data forensic and cross-reference analy-
sis. In Proceedings of the 2011 ACM Symposium on Applied Computing,
pages 185–190. ACM, 2011.
[18] M. Huber, M. Mulazzani, M. Leithner, S. Schrittwieser, G. Wondracek,
and E. Weippl. Social snapshots: Digital forensics for online social net-works. Proceedings of the 27th Annual Computer Security Applications
Conference (ACSAC), 2011.
[19] M. Huber, M. Mulazzani, E. Weippl, G. Kitzler, and S. Goluch. Friend-
in-the-middle attacks: Exploiting social networking sites for spam. IEEE
Internet Computing: Special Issue on Security and Privacy in Social
Networks, 5 2011. Pre Print.
[20] J. Marques. Firefox extensions. Online at http://codebutler.com/
firesheep.
[21] S. Morrissey. iOS Forensic Analysis. Apress, 2010.
[22] S. Teelink and R. Erbacher. Improving the computer forensic analysis
process through visualization. Communications of the ACM, 49(2):71–
75, 2006.
[23] S. Wasserman. Social network analysis: Methods and applications. Cam-
bridge university press, 1994.You can also read
